CN114176549A - Fetal heart rate signal data enhancement method and device based on generative countermeasure network - Google Patents
Fetal heart rate signal data enhancement method and device based on generative countermeasure network Download PDFInfo
- Publication number
- CN114176549A CN114176549A CN202111590892.1A CN202111590892A CN114176549A CN 114176549 A CN114176549 A CN 114176549A CN 202111590892 A CN202111590892 A CN 202111590892A CN 114176549 A CN114176549 A CN 114176549A
- Authority
- CN
- China
- Prior art keywords
- data
- layer
- fhr
- discriminator
- confrontation network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 28
- 210000002458 fetal heart Anatomy 0.000 title claims abstract description 19
- 238000009826 distribution Methods 0.000 claims abstract description 25
- 238000004088 simulation Methods 0.000 claims abstract description 21
- 238000013528 artificial neural network Methods 0.000 claims abstract description 3
- 230000006870 function Effects 0.000 claims description 51
- 238000013527 convolutional neural network Methods 0.000 claims description 14
- 238000012549 training Methods 0.000 claims description 13
- 230000004913 activation Effects 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 5
- 238000003475 lamination Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 3
- 238000004590 computer program Methods 0.000 claims description 2
- 238000010276 construction Methods 0.000 claims description 2
- 238000003860 storage Methods 0.000 claims description 2
- 239000013598 vector Substances 0.000 claims description 2
- 238000013256 Gubra-Amylin NASH model Methods 0.000 abstract description 8
- 210000003754 fetus Anatomy 0.000 description 8
- 238000004422 calculation algorithm Methods 0.000 description 6
- 230000001575 pathological effect Effects 0.000 description 6
- 238000003745 diagnosis Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000002708 enhancing effect Effects 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 230000009429 distress Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000036541 health Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 208000009119 Giant Axonal Neuropathy Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000012631 diagnostic technique Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000001605 fetal effect Effects 0.000 description 1
- 201000003382 giant axonal neuropathy 1 Diseases 0.000 description 1
- 235000003642 hunger Nutrition 0.000 description 1
- 230000003902 lesion Effects 0.000 description 1
- 210000004185 liver Anatomy 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000035935 pregnancy Effects 0.000 description 1
- 230000037351 starvation Effects 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/024—Detecting, measuring or recording pulse rate or heart rate
- A61B5/02411—Detecting, measuring or recording pulse rate or heart rate of foetuses
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B2503/00—Evaluating a particular growth phase or type of persons or animals
- A61B2503/02—Foetus
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Pathology (AREA)
- Physiology (AREA)
- Computing Systems (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- Animal Behavior & Ethology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Surgery (AREA)
- Medical Informatics (AREA)
- Heart & Thoracic Surgery (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Cardiology (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Pregnancy & Childbirth (AREA)
- Gynecology & Obstetrics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Fuzzy Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a fetal heart rate signal data enhancement method and device based on a generative confrontation network. Adopting a micro-step convolution and step convolution design generator and a discriminator to construct a GAN model based on a deep convolution neural network structure; measuring the distance between the real acquired FHR sample and the simulation data by adopting a Wasserstein distance with a gradient penalty, and optimizing a model objective function; establishing an auxiliary classifier based on class constraint, and performing reverse updating operation on model parameters of the GAN model by using the auxiliary classifier; and inputting the acquired incomplete FHR signal, the noise data meeting the standard normal distribution and the class label of the real FHR sample into the optimized GAN model to generate simulated FHR data so as to realize data enhancement of the fetal heart rate signal.
Description
Technical Field
The invention relates to the technical field of modern medical signal processing, in particular to a fetal heart rate signal data enhancement method and device based on a generative confrontation network.
Background
When the pregnant woman has problems in the late pregnancy and delivery, accurate diagnosis and timely treatment of the health condition of the fetus according to the existing information of the pregnant woman, the pregnant woman and the fetus are particularly important. In clinical practice, Fetal Heart Rate (FHR) monitoring is a widely used prenatal and intrapartum diagnostic technique used by trained clinical care personnel to assess the health of a fetus and to detect a dangerous fetus early so that appropriate and timely action can be taken to prevent further damage to the fetus and mother. However, any examination relying on human interpretation in clinic is affected by the clinical experience and ability level of doctors, and problems of poor accuracy and consistency exist generally, and the human information interpretation is highly controversial. Therefore, it is important to construct an automatic intelligent medical auxiliary diagnosis tool to assist the clinician in accurately diagnosing the pathological condition of the fetus.
One of the difficulties in implementing intelligent medical auxiliary diagnosis is that sufficient training data, especially an auxiliary diagnosis algorithm using deep learning as a main tool, often requires a large amount of data to learn model parameters, and the effectiveness thereof depends on the number of labeled samples to a great extent. In the medical field, the number of healthy fetuses is far greater than the number of abnormal distress cases, and the number of patients who can participate in the sample data collection experiment is much smaller. Data starvation and class imbalance significantly hinder the performance of deep learning algorithms. Therefore, by using the data enhancement algorithm to generate the medical data of a specific class, acquiring more high-quality few types of sample data helps to alleviate the above problem.
Data enhancement, that is, new simulation data is generated by using existing data, and is common in the field of computer vision, for example, turning and rotating processing of various angles is widely used in deep learning tasks of medical images, such as liver lesion classification. On the other hand, in the one-dimensional time domain, the sample data is often expanded by means of noise and time warping, but such algorithms may cause unnecessary changes in physiological signals, thereby affecting the reliability of subsequent classification. Therefore, a data enhancement method capable of generating one-dimensional data accurately and truly is required.
In view of the above, it is desirable to provide a data enhancement algorithm for an FHR signal, which achieves expansion of a few types of samples, and further provides sufficient high-quality sample data for implementation of an intelligent medical auxiliary diagnosis algorithm for fetal status based on the FHR signal.
Disclosure of Invention
An object of the present invention is to solve the above technical problems, and provide a method for enhancing fetal heart rate signal data based on a Generative Adaptive Network (GAN), which uses a deep convolutional neural network model, fuses Wasserstein distances with gradient penalties, and generates a simulated FHR signal of any length with an auxiliary classifier based on class constraints.
In order to solve the problems, the invention adopts the following technical scheme:
the invention discloses a method for enhancing FHR signal data, which comprises the following steps:
s1, constructing a generative confrontation network model based on the deep convolutional neural network structure, which comprises the following steps: realizing up-sampling based on a micro-step convolution function, realizing down-sampling by adopting step convolution, respectively designing a generator and a discriminator, and establishing a generating type confrontation network model based on a deep convolution neural network;
s2, Walserstein distance with gradient penalty, optimizing an objective function, including: and measuring the distance between the two distributions of real data and simulation data based on the Wasserstein distance with the gradient penalty, and optimizing the GAN model constructed in the previous step. Compared with the traditional GAN model, the Jensen-Shannon distance, namely the JS distance, is usually adopted in the measurement of the loss function of the generator to minimize the fitting degree of the generated distribution and the real distribution, but when the two distributions are not overlapped or are overlapped extremely small, the problem of gradient disappearance is caused, and the loss function optimization scheme provided by the invention solves the unstable problem of the GAN model in the training process;
s3, establishing an auxiliary classifier based on class constraint, ensuring the diversity of simulation data and solving the problem of mode collapse of the traditional generation type countermeasure network model; the method comprises the following steps: and step convolution is adopted as a down-sampling strategy, an auxiliary classifier based on class constraint is established, the diversity of simulation data is guaranteed, and the problem of mode collapse of the traditional GAN model is solved.
S4, optimizing the objective function by using the auxiliary classifier to realize the reverse updating operation of the model parameters of the generative confrontation network model; the method comprises the following steps: and the generator takes the random normal distribution noise data cut off to a certain range and the class label of the real FHR as input, and outputs the simulation FHR data and the discrimination result of the auxiliary classifier on the simulation data. The discriminator makes a judgment result according to the real FHR sample and the generated simulation FHR data, and the auxiliary classifier obtains a class label according to the input simulation data and carries out reverse updating operation on all parameters.
Another object of the present invention is to provide a fetal heart rate signal data enhancement device based on a generative confrontation network, comprising:
the main network construction module is used for constructing a generating type confrontation network model based on a deep convolutional neural network structure;
the optimization module is used for optimizing a generative confrontation network model objective function based on a deep convolutional neural network structure;
the auxiliary classifier module is used for constructing an auxiliary classifier based on class constraint;
and the model parameter updating module is used for optimizing the model objective function by using the auxiliary classifier based on the category constraint to realize the reverse updating of the main network model parameters.
It is a further object of the present invention to provide a computer-readable storage medium having stored thereon a computer program which, when executed in a computer, causes the computer to perform the above-mentioned method.
It is a further object of the present invention to provide a computing device comprising a memory having stored therein executable code and a processor that, when executing the executable code, implements the method described above.
The invention has the beneficial effects that:
(1) in the confrontation training of the GAN model, the loss function is directly related to the convergence condition of model training, and the Wasserstein distance with gradient punishment is provided to measure the fitting degree of two distributions and optimize the calculation of the loss function. On the basis of the traditional Wasserstein distance, a penalty item which enables the gradient to be associated with a first-order Lipschitz constant is added, so that the discriminator cannot concentrate the discrimination results of most data on a threshold boundary, the overfitting phenomenon of the discriminator is prevented, the problems of gradient explosion and disappearance are avoided, and the training stability is improved.
(2) An auxiliary classifier is added to realize small sample generation based on class constraint, and the problem of mode collapse is solved. Mode collapse, that is, most of the current small sample expansion models only tend to generate some kind or some kinds of data, and the samples generated by simulation lack diversity. However, the FHR signals of clinically healthy fetuses and abnormal distress cases are very different in waveform representation forms, and when such a one-to-many mapping relationship occurs, the existing GANs model cannot meet the requirement for fast generation of two types of simulation data, and the model structure must be retrained and adjusted, which brings huge model training overhead. The invention adds a category constraint condition to the original GANS model, and realizes the conversion generation of multi-category samples in one model by adding an auxiliary classifier.
Drawings
FIG. 1 is a flow chart of an embodiment;
FIG. 2 is a schematic diagram of the structure of a generator, discriminator and auxiliary classifier;
FIG. 3 is a schematic diagram of an example of a normal FHR sample and its small sample expansion from a starting database; wherein (a) normal fetal heart rate samples from the CTU-UHB database, (a) simulation generated normal fetal heart rate data;
FIG. 4 is a schematic diagram of an example of a pathological FHR sample and its small sample expansion from a starting database; wherein (a) the pathological fetal heart rate sample from the CTU-UHB database and (a) the generated pathological fetal heart rate data are simulated.
Detailed Description
The technical scheme of the invention is further specifically described by the following embodiments and the accompanying drawings.
A method for enhancing FHR signal data in this embodiment, as shown in fig. 1 and fig. 2, includes the following steps:
s1, constructing a generative confrontation network model based on the deep convolutional neural network structure; as shown in fig. 2
The generative confrontation network model based on the deep convolutional neural network structure comprises a generator and a discriminator;
1) the input of the generator is noise data meeting standard normal distribution and a class label of a real FHR sample, wherein the class label of a normal FHR sample is marked as 0, and the class label of a pathological FHR sample is marked as 1; outputting simulated FHR data with the same size as the input noise data;
for the generator, a micro-step convolution function is adopted to realize up-sampling; firstly, fusing and mapping noise data and label data into vectors with specified sizes; then, 6 deconvolution operations are sequentially carried out, wherein each operation comprises deconvolution, batch normalization and activation function processing, and simulated FHR data with the same size as the input noise data are obtained;
2) the input of the discriminator is a real FHR sample and simulated FHR data generated by the generator, and the output is an discrimination result of the category to which the simulated FHR data belongs;
for the discriminator, the step convolution is adopted to realize down sampling; the input data goes through 5 convolutional layer operations in sequence: the output of the first 4 convolutions is processed by batch normalization, and LeakyReLU is adopted as an activation function to increase the nonlinearity of the network; then, the output of the 5 th layer is connected with a Flatten layer for flattening operation, and the multidimensional input is subjected to one-dimensional operation; then, fitting processing is carried out on the fully connected layer and the Dropout layer, and a LeakyReLU activation function layer is connected; finally, entering a full connection layer with an activation function of Sigmod to judge the consistency of the simulation data and the real sample;
3) the initial objective function was constructed as shown in equation (1) below:
wherein E (-) denotes an expectation value, G denotes a generator, D denotes a discriminator, PdataAnd PzRespectively representing the distribution of the real and the generated FHR data, V (G, D) representing a cross-entropy function of two classes, the noise data z in the input of the generator G obeying the data distribution pzThe output is G (z), the real sample x in the input item of the discriminator D obeys the distribution PdataThe output is d (x), and the final purpose of formula (1) is to minimize the relative entropy between the simulation sample and the real data, i.e. KL divergence;
s2, optimizing the formula (1) by the Wasserstein distance with gradient penalty; the method comprises the following steps:
2-1 uses Wasserstein distance to measure the distance between two distributions:
wherein gamma represents combined distribution, gamma-pi (P)data,Pz) I.e. represents PdataAnd PzA set of all possible joint distributions combined;
2-2, introducing a Lipschitz continuity condition, and designing a gradient penalty method to carry out optimization calculation on the formula (2):
(ii) data distribution P for true and false samples according to the following equation (3)dataAnd PzRandom interpolation sampling is carried out to generate a new sample
Where ξ is a random number between [0,1 ];
② calculating gradient of discriminator by random interpolation sampling
Setting the first-order Lipschitz constant of the discriminator to be K, and establishing
And K, realizing gradient penalty term solution:
wherein λ is a parameter that adjusts the magnitude of the gradient penalty term;
to represent
FHR data distribution of (a);
③ expressing the loss function of the discriminator in the objective function of equation (1) as:
wherein the experience value of K is 1;
s3, establishing an auxiliary classifier based on class constraint, ensuring the diversity of simulation data and solving the problem of mode collapse of the traditional generation type countermeasure network model; the method comprises the following steps:
the auxiliary classifier comprises an input layer, a middle layer and an output layer; the input layer receives the simulated FHR data generated by the generator; the intermediate layer adopts four first-step coiled layer, one second-step coiled layer, a full connection layer, a Dropout layer and a LeakyReLU activation function layer which are sequentially cascaded; the first step format convolution layer comprises a convolution layer, a batch normalization processing layer and a LeakyReLU activation function layer; the second-step scroll lamination layer comprises a scroll lamination layer and a Flatten layer;
the classifier of the output layer adopts a Softmax classifier and is used for predicting a class label to which FHR data generated by simulation belongs;
s4: optimizing the objective function by using an auxiliary classifier to realize reverse updating operation on the model parameters of the generative confrontation network model; as fig. 2 specifically shows:
the objective function of the optimized generative confrontation network model comprises a log-likelihood function L of the probability of the correct source of the simulation datazAnd a log-likelihood function L of the probability of correct class labelcAs shown in formulas (6) to (7);
in conjunction with equations (5) - (7), the objective of the overall network training is to optimize the discriminator to maximize the loss function value Lz+Lc+ Ω, optimize generator G to maximize the loss function value Lc-Lz；
Wherein L isDValue of loss function, L, representing discriminatorGRepresents the loss function value of the generator G;
s5: and generating simulated FHR data by using the generated confrontation network model based on the deep convolutional neural network structure optimized in the step S4 so as to realize the data enhancement of the fetal heart rate signal.
Setting training parameters: an Adam optimizer is used in the training process, and the Adam optimizer relates to three hyper-parameters which are respectively set to be beta1＝0.9,β2＝0.999,ε＝10-8(ii) a The number of critical iterations for each generator iteration is set to 5. In addition, the optimal batch size, initial learning rate initial spare rate and training iteration cycle epochs are set to be 1000 and 10 respectively according to the loss rate curve representation reflected by the loss function in the training process-3,40. And training and saving the model for generating simulation data of FHR.
The data enhancement experiment of the fetal heart rate signal is carried out by utilizing the generated confrontation network model based on the deep convolutional neural network structure after the optimization of the invention by combining an open source database provided by Czech Technical University-University in Hospital in Brno, CTU-UHB of Czech Technical University Hospital. 1 group of normal FHR samples and pathological FHR samples were randomly selected from CTU-UHB for small sample expansion, and the results shown in FIGS. 3-4 were obtained.
The above embodiments are not intended to limit the present invention, and the present invention is not limited to the above embodiments, and all embodiments are within the scope of the present invention as long as the requirements of the present invention are met.
Claims (9)
1. A fetal heart rate signal data enhancement method based on a generative confrontation network is characterized by comprising the following steps:
s1: constructing a generating type confrontation network model based on a deep convolution neural network structure;
the generative confrontation network model based on the deep convolutional neural network structure comprises a generator and a discriminator;
1) the input of the generator is noise data meeting standard normal distribution and a class label of a real FHR sample; outputting simulated FHR data with the same size as the input noise data;
2) the input of the discriminator is a real FHR sample and simulated FHR data generated by the generator, and the output is an discrimination result of the category to which the simulated FHR data belongs;
3) the initial objective function was constructed as shown in equation (1) below:
wherein E (-) denotes an expectation value, G denotes a generator, D denotes a discriminator, PdataAnd PzRespectively representing the distribution of the real and the generated FHR data, V (G, D) representing a cross-entropy function of two classes, the noise data z in the input of the generator G obeying the data distribution pzThe output is G (z), the real sample x in the input item of the discriminator D obeys the distribution PdataThe output is d (x), and the final purpose of formula (1) is to minimize the relative entropy between the simulation sample and the real data, i.e. KL divergence;
s2: optimizing an objective function by using the Wasserstein distance with gradient penalty; the method comprises the following steps:
2-1 uses Wasserstein distance to measure the distance between two distributions:
wherein gamma represents combined distribution, gamma-pi (P)data,Pz) I.e. represents PdataAnd PzA set of all possible joint distributions combined;
2-2, introducing a Lipschitz continuity condition, and designing a gradient penalty method to optimize the formula (2);
s3: establishing an auxiliary classifier based on class constraint;
the auxiliary classifier comprises an input layer, a middle layer and an output layer; the input layer receives the simulated FHR data generated by the generator; the intermediate layer adopts four first-step coiled layer, one second-step coiled layer, a full connection layer, a Dropout layer and a LeakyReLU activation function layer which are sequentially cascaded; the first step format convolution layer comprises a convolution layer, a batch normalization processing layer and a LeakyReLU activation function layer; the second-step scroll lamination layer comprises a scroll lamination layer and a Flatten layer;
s4: optimizing the objective function by using an auxiliary classifier to realize reverse updating operation on the model parameters of the generative confrontation network model;
s5: and generating simulated FHR data by using the generated confrontation network model based on the deep convolutional neural network structure optimized in the step S4 so as to realize the data enhancement of the fetal heart rate signal.
2. The method as claimed in claim 1, wherein the generative confrontation network based generative heart rate signal data enhancement method based on the generative confrontation network structure is characterized in that the generator in the generative confrontation network model based on the deep convolutional neural network structure adopts a micro-step convolution function to realize the upsampling: firstly, fusing and mapping noise data and label data into vectors with specified sizes; and then sequentially carrying out 6 deconvolution operations, wherein each deconvolution operation comprises a deconvolution layer, batch normalization and activation function processing, and obtaining simulated FHR data with the same size as the input noise data.
3. The method as claimed in claim 1, wherein the step-size convolution is adopted by the discriminator in the generative confrontation network model based on the deep convolutional neural network structure to realize the down-sampling; the input data goes through 5 convolutional layer operations in sequence: the output of the first 4 convolutions is processed by batch normalization, and LeakyReLU is used as an activation function; then, the output of the 5 th layer is connected with a Flatten layer for flattening operation, and the multidimensional input is subjected to one-dimensional operation; then, fitting processing is carried out on the fully connected layer and the Dropout layer, and a LeakyReLU activation function layer is connected; and finally, entering a full connection layer with an activation function of Sigmod to judge the consistency of the simulation data and the real sample.
4. The fetal heart rate signal data enhancement method based on the generative countermeasure network as claimed in claim 1, wherein the step 2-2 is specifically:
(ii) data distribution P for true and false samples according to the following equation (3)dataAnd PzA random interpolation sampling is performed and,generating a new sample
Where ξ is a random number between [0,1 ];
② calculating gradient of discriminator by random interpolation sampling
Setting the first-order Lipschitz constant of the discriminator to be K, and establishing
And K, realizing gradient penalty term solution:
wherein λ is a parameter that adjusts the magnitude of the gradient penalty term;
to represent
FHR data distribution of (a);
③ expressing the loss function of the discriminator in the objective function of equation (1) as:
wherein K is 1.
5. The method as claimed in claim 1, wherein the classifier of the output layer of the auxiliary classifier based on class constraint adopts a Softmax classifier for predicting the class label to which the simulated generated FHR data belongs.
6. The fetal heart rate signal data enhancement method based on the generative confrontation network as claimed in claim 1, wherein the step S4 is specifically:
the objective function of the optimized generative confrontation network model comprises a log-likelihood function L of the probability of the correct source of the simulation datazAnd a log-likelihood function L of the probability of correct class labelcAs shown in formulas (6) to (7);
in conjunction with equations (5) - (7), the objective of the overall network training is to optimize the discriminator to maximize the loss function value Lz+Lc+ Ω, optimize generator G to maximize the loss function value Lc-Lz；
Wherein L isDValue of loss function, L, representing discriminatorGRepresenting the loss function value of generator G.
7. A fetal heart rate signal data enhancement device based on a generative confrontation network, comprising:
the main network construction module is used for constructing a generating type confrontation network model based on a deep convolutional neural network structure;
the optimization module is used for optimizing a generative confrontation network model objective function based on a deep convolutional neural network structure;
the auxiliary classifier module is used for constructing an auxiliary classifier based on class constraint;
and the model parameter updating module is used for optimizing the model objective function by using the auxiliary classifier based on the category constraint to realize the reverse updating of the main network model parameters.
8. A computer-readable storage medium, on which a computer program is stored which, when executed in a computer, causes the computer to carry out the method of any one of claims 1-6.
9. A computing device comprising a memory having executable code stored therein and a processor that, when executing the executable code, implements the method of any of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111590892.1A CN114176549B (en) | 2021-12-23 | 2021-12-23 | Fetal heart rate signal data enhancement method and device based on generation type countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111590892.1A CN114176549B (en) | 2021-12-23 | 2021-12-23 | Fetal heart rate signal data enhancement method and device based on generation type countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114176549A true CN114176549A (en) | 2022-03-15 |
| CN114176549B CN114176549B (en) | 2024-04-16 |
Family
ID=80605971
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111590892.1A Active CN114176549B (en) | 2021-12-23 | 2021-12-23 | Fetal heart rate signal data enhancement method and device based on generation type countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114176549B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114530250A (en) * | 2022-04-24 | 2022-05-24 | 广东工业大学 | Wearable blood glucose detection method and system based on data enhancement and storage medium |
| CN116680637A (en) * | 2023-08-02 | 2023-09-01 | 北京世纪慈海科技有限公司 | Construction method and device of sensing data analysis model of community-built elderly people |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918999A (en) * | 2019-01-22 | 2019-06-21 | 西安交通大学 | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database |
| CN111476294A (en) * | 2020-04-07 | 2020-07-31 | 南昌航空大学 | Zero sample image identification method and system based on generation countermeasure network |
| CN111539467A (en) * | 2020-04-17 | 2020-08-14 | 北京工业大学 | GAN network architecture and method for data augmentation of medical image data set based on generation of countermeasure network |
| CN112784930A (en) * | 2021-03-17 | 2021-05-11 | 西安电子科技大学 | CACGAN-based HRRP identification database sample expansion method |
| CN113139916A (en) * | 2021-04-15 | 2021-07-20 | 哈尔滨工程大学 | Underwater sonar simulation image generation and data expansion method based on generation type countermeasure network |
| CN113269314A (en) * | 2021-06-16 | 2021-08-17 | 福州大学 | New energy power generation scene data migration method based on generation countermeasure network |
-
2021
- 2021-12-23 CN CN202111590892.1A patent/CN114176549B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918999A (en) * | 2019-01-22 | 2019-06-21 | 西安交通大学 | Based on the mechanical equipment fault intelligent diagnosing method for generating model under a kind of Small Sample Database |
| CN111476294A (en) * | 2020-04-07 | 2020-07-31 | 南昌航空大学 | Zero sample image identification method and system based on generation countermeasure network |
| CN111539467A (en) * | 2020-04-17 | 2020-08-14 | 北京工业大学 | GAN network architecture and method for data augmentation of medical image data set based on generation of countermeasure network |
| CN112784930A (en) * | 2021-03-17 | 2021-05-11 | 西安电子科技大学 | CACGAN-based HRRP identification database sample expansion method |
| CN113139916A (en) * | 2021-04-15 | 2021-07-20 | 哈尔滨工程大学 | Underwater sonar simulation image generation and data expansion method based on generation type countermeasure network |
| CN113269314A (en) * | 2021-06-16 | 2021-08-17 | 福州大学 | New energy power generation scene data migration method based on generation countermeasure network |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114530250A (en) * | 2022-04-24 | 2022-05-24 | 广东工业大学 | Wearable blood glucose detection method and system based on data enhancement and storage medium |
| CN116680637A (en) * | 2023-08-02 | 2023-09-01 | 北京世纪慈海科技有限公司 | Construction method and device of sensing data analysis model of community-built elderly people |
| CN116680637B (en) * | 2023-08-02 | 2023-11-03 | 北京世纪慈海科技有限公司 | Construction method and device of sensing data analysis model of community-built elderly people |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114176549B (en) | 2024-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107680676B (en) | Gestational diabetes prediction method based on electronic medical record data drive | |
| Lu et al. | Prediction of fetal weight at varying gestational age in the absence of ultrasound examination using ensemble learning | |
| Abouelyazid et al. | Machine Learning-Assisted Approach for Fetal Health Status Prediction using Cardiotocogram Data | |
| Petrozziello et al. | Deep learning for continuous electronic fetal monitoring in labor | |
| CN114176549B (en) | Fetal heart rate signal data enhancement method and device based on generation type countermeasure network | |
| CN108847285B (en) | Down syndrome screening method for pre-pregnancy and mid-pregnancy based on machine learning | |
| EP3964136A1 (en) | System and method for guiding a user in ultrasound assessment of a fetal organ | |
| WO1999009507A1 (en) | Methods for selecting, developing and improving diagnostic tests for pregnancy-related conditions | |
| Rachmatullah et al. | Convolutional neural network for semantic segmentation of fetal echocardiography based on four-chamber view | |
| CN113610118A (en) | Fundus image classification method, device, equipment and medium based on multitask course learning | |
| Spairani et al. | A deep learning mixed-data type approach for the classification of FHR signals | |
| CN116306783A (en) | FHR data enhancement method based on LSTM-DCGAN model | |
| CN113469218A (en) | Electrocardiosignal critical value processing method, system and medium based on deep learning | |
| Liang et al. | Fully automatic classification of cardiotocographic signals with 1D-CNN and bi-directional GRU | |
| Kapila et al. | Optimizing fetal health prediction: Ensemble modeling with fusion of feature selection and extraction techniques for cardiotocography data | |
| CN112992353A (en) | Method and device for accurately predicting due date, computer equipment and storage medium | |
| CN116563224B (en) | Image histology placenta implantation prediction method and device based on depth semantic features | |
| WO2023024524A1 (en) | Fetal ultrasonic radiomics feature-based chromosomal abnormality prediction model construction method and diagnosis device | |
| CN111148999A (en) | Method and computer program for predicting neonatal bilirubin levels | |
| Marandi et al. | A Convolution Neural Network-based Prediction of Fetal Health | |
| CN114287900A (en) | Missing value filling device and method for fetal heart rate signals | |
| Gaddam et al. | Fetal Abnormality Detection: Exploring Trends Using Machine Learning and Explainable AI | |
| Chidambaram et al. | Comparative Analysis of Ensemble Learning Methods for Enhancing Fetal Health Prediction Using Cardiotocography | |
| Arman et al. | Intracranial hemorrhage classification from ct scan using deep learning and bayesian optimization | |
| Sahana et al. | Prediction of prenatal infants birth weight in pregnancies |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |